← All docs

Discovery

Discovery surfaces structure hiding in your particle data — patterns that emerge from feature similarity rather than from labels you’ve already applied.

Atlases and constellations

A Discovery job reduces particle features to a navigable map called an

. Within it, a clustering algorithm groups similar particles into s. The astronomy metaphor is deliberate: a constellation *might* reflect real biological structure, or it might be a coincidence of the feature space. You decide which.

The constellation lifecycle

As you review a constellation it moves through states:

  • Unreviewed — fresh algorithm output.
  • Named — you’ve inspected it and given it a descriptive name.
  • Mapped — it corresponds to an existing class; you link it to a collection.
  • Promoted — it represents a new class; you create a collection seeded with its particles.
  • Dismissed — not biologically meaningful (an artifact or noise).

Mapping links discovery to knowledge you already have; promotion turns a discovery into new knowledge.

Discovery vs Explorer

Discovery vs Explorer. Explorer visualizes what you already have, instantly and client-side. Discovery runs heavier, asynchronous compute jobs to find groupings you hadn’t defined. Explorer answers “what’s in my data?”; Discovery answers “what patterns are hiding in it?”

Quick start

  1. Choose a scope to embed (a series, collection, or compendium).
  2. Generate an and let the job run.
  3. Review each — name, map, promote, or dismiss it.